SoftBank builds Japan’s most powerful AI supercomputer with NVIDIA Blackwell for broader sovereign AI initiatives, announces plans for Grace Blackwell NVIDIA AI Aerial enables SoftBank to build the world’s first live 5G AI-RAN, SoftBank uses NVIDIA AI Enterprise to create billions of dollars in new revenue opportunities for the global telecommunications industry Build a marketplace and meet the national demand for local, secure AI computing
NVIDIA AI Summit Japan — NVIDIA today aims to accelerate Japan’s sovereign AI efforts, furthering global technology leadership and unlocking billions of dollars in AI revenue opportunities for communications providers around the world announced a series of collaborations with SoftBank Corp.
In a keynote speech at NVIDIA AI Summit Japan, NVIDIA founder and CEO Jensen Huang said that SoftBank is using the NVIDIA Blackwell platform to build Japan’s most powerful AI supercomputer, and that NVIDIA will use NVIDIA for its next supercomputer. announced plans to use the Grace Blackwell platform.
Additionally, NVIDIA revealed that SoftBank successfully conducted the world’s first combined AI and 5G communication network pilot using the NVIDIA AI Aerial accelerated computing platform. This is a computing breakthrough that opens up potentially billions of dollars worth of AI revenue streams for carriers.
NVIDIA and SoftBank also announced that, using NVIDIA AI Enterprise software, SoftBank aims to build an AI marketplace where SoftBank can meet the demand for local, secure AI computing. This new service, which supports AI training and edge AI inference, positions SoftBank as Japan’s AI grid, driving new business opportunities for the creation, distribution, and use of AI services across Japanese industries, consumers, and enterprises. .
“Japan has a long history of leading technological innovations that impact the world,” Huang said. “SoftBank’s significant investment in NVIDIA’s full-stack AI, Omniverse, and 5G AI-RAN platforms will allow Japan to jump into the AI industrial revolution and become a global leader, with power across the telecommunications, transportation, robotics, and healthcare industries. We are driving a new era of growth” in a way that will greatly benefit humanity in the age of AI. ”
Junichi Miyagawa, President and CEO of SoftBank, said, “Countries and regions around the world are accelerating the introduction of AI in order to achieve social and economic growth, and society is undergoing major changes.” “Through our long-standing collaboration with NVIDIA, SoftBank is leading this transformation from the front, leveraging our extremely powerful AI infrastructure and new distributed AI-RAN solution to reinvent 5G networks for AI.” With AITRAS, we will accelerate innovation across the country and around the world. ”
SoftBank first acquires Mr. Blackwell, plans to acquire Grace Blackwell
SoftBank will receive the world’s first NVIDIA DGX™ B200 system, which will serve as a building block for the new NVIDIA DGX SuperPOD™ supercomputer.
SoftBank plans to use DGX SuperPOD, which is powered by Blackwell, not only for its own generative AI development and AI-related business, but also for the business of universities, research institutes, and companies across Japan.
Once completed, SoftBank’s DGX SuperPOD is expected to be Japan’s highest performance to date. Featuring NVIDIA AI Enterprise software and NVIDIA Quantum-2 InfiniBand networking, it’s also ideal for developing language models at scale.
In addition to DGX SuperPOD, SoftBank plans to build another NVIDIA-accelerated supercomputer to run highly compute-intensive workloads. Initial plans for the supercomputer are based on the NVIDIA Grace Blackwell platform design, which features an NVIDIA GB200 NVL72 multi-node liquid-cooled rack-scale system that combines NVIDIA Blackwell GPUs and power-efficient Arm-based NVIDIA Grace™ CPUs.
AI-RAN achieves new milestone
SoftBank is working closely with NVIDIA on a technology milestone to develop a new type of telecommunications network that can run AI and 5G workloads simultaneously, known in the industry as Artificial Intelligence Radio Access Network (AI-RAN). Achieved.
This new type of infrastructure has broad ecosystem support from the telecom industry as it provides carriers with the ability to transform base stations from cost centers to AI revenue-generating assets.
Through an outdoor trial conducted in Kanagawa Prefecture, SoftBank demonstrates that its NVIDIA-accelerated AI-RAN solution can achieve carrier-grade 5G performance and leverage network excess capacity to concurrently run AI inference workloads I did.
Traditional communication networks are designed to handle peak loads, and on average only use one-third of their capacity. The common computing capabilities provided by AI-RAN are expected to give telcos the opportunity to monetize the remaining two-thirds of capacity for AI inference services.
NVIDIA and SoftBank estimate that for every $1 of capital investment carriers invest in new AI-RAN infrastructure, they can earn approximately $5 in AI inference revenue. (1) Considering operating costs and capital investment costs, SoftBank estimates that it can achieve the highest profit margin. 219% for each AI-RAN server added to your infrastructure. (2)
Perform real-world inference with AI-RAN
In this trial, SoftBank used NVIDIA AI Enterprise to build real-world AI inference applications, including remote support for self-driving cars, robotics control, and automated multimodal search generation at the edge. All inference workloads ran optimally on SoftBank’s AI-RAN network.
SoftBank’s fully software-defined 5G radio stack is optimized for NVIDIA’s AI computing platform and includes SoftBank-enhanced L1 software based on NVIDIA Aerial™ CUDA®-accelerated RAN libraries. Included. SoftBank plans to incorporate the NVIDIA Aerial RAN Computer-1 system into future solutions, which it estimates can consume 40% less power than traditional 5G network infrastructure (3).
NVIDIA and SoftBank partners who contributed to trials of SoftBank’s AI-RAN solution include Fujitsu and Red Hat.
Match supply and demand
SoftBank uses NVIDIA AI Enterprise Serverless because AI-RAN solutions must dynamically spin up or down compute based on supply and demand without compromising carrier-grade performance in real time. We aim to build an ecosystem that connects the demand and supply of AI technology. Orchestrator of application programming interfaces and their in-house development. This will enable SoftBank to dispatch external AI inference jobs to AI-RAN servers when computing resources are available to provide localized, low-latency, secure inference services.
“Moving from a single-purpose AI-RAN network to a multi-purpose AI-RAN network can deliver five times the return per dollar of capital investment,” said Ronnie Vasishta, senior vice president of communications at NVIDIA. states. “SoftBank’s live field trial validates the technology’s feasibility, performance, and economics, and is a major step toward commercialization of AI-RAN.”
“SoftBank’s AITRAS is the first AI-RAN solution developed through a five-year collaboration with NVIDIA. “We increase communication efficiency by running high-density cells on a single GPU server,” said Ryuji Wakukawa, vice president and director of the Advanced Technology Research Institute at SoftBank. “We are confident that this AI-driven innovation, AITRAS, will pave the way for new business models in communications and will be a key element in the transformation of mobile operators.”
Learn more about NVIDIA solutions for AI-RAN.
